In an effort to advance early detection of colon cancer, Sioux Falls, S.D.-based Sanford Health has developed a predictive AI model designed to identify individuals at elevated risk for the disease.
The model, driven by machine learning and leveraging electronic medical records, assesses risk factors beyond the standard metrics, helping physicians prioritize patients who may require more urgent screening.
"Our goal is to ensure that every patient receives top-quality care, including essential screenings for colorectal cancer," Jeremy Cauwels, MD, chief medical officer at Sanford Health, told Becker's. "This model helps us identify those at heightened risk with greater accuracy, allowing us to effectively reach out to individuals who might otherwise go untested."
Sanford Health's model not only focuses on traditional risk factors such as age, family history, and lifestyle, but also incorporates lesser-known variables that may be overlooked in conventional screening. John Bassett, MD, chair of the gastroenterology department at Sanford Fargo (N.D.) Medical Center, told Becker's the model considers 85 variables, allowing the system to stratify risk more effectively.
"With this model, we can narrow down those within an average-risk pool to identify higher and lower risk levels, allowing us to allocate resources like colonoscopies to those most in need," Dr. Bassett said.
A significant motivation for the development of this model arose from the recent recommendation by the U.S. Preventive Services Task Force to lower the age for routine colon cancer screenings from 50 to 45.
"This change added 100,000 newly eligible patients to our system alone, challenging our limited resources," Dr. Bassett said. "The AI model helps us meet this demand by identifying those at highest risk first."
Developed in collaboration with Sanford Health’s data analytics and biostatistics teams, the model draws from a dataset of more than 450,000 individuals ages 45 to 80.
"All data used in training and analysis is de-identified, allowing our teams to work without knowing any individual patient's identity," Dr. Cauwels said. "This is a fundamental priority, especially as we consider scaling this model or collaborating with other health systems."
As Sanford Health continues to pilot the model, its leaders hope to build an evidence base supporting its efficacy, with plans to publish findings and implement a systemwide rollout. Both Dr. Bassett and Dr. Cauwels are optimistic about the future of this AI-driven approach, with aspirations that it will become a standard component of cancer screening nationwide.
"The potential of this model to prioritize high-risk patients could reduce wait times and improve early detection rates," Dr. Cauwels said. "It's an exciting step forward that underscores Sanford Health's commitment to proactive, data-driven healthcare."